Transcript Tessellation OS Architecting Systems Software in a ManyCore World John Kubiatowicz UC Berkeley [email protected] Uniprocessor Performance (SPECint) Performance (vs.

Tessellation OS
Architecting Systems Software in a
ManyCore World
John Kubiatowicz
UC Berkeley
[email protected]
Uniprocessor Performance (SPECint)
Performance (vs. VAX-11/780)
10000
From Hennessy and Patterson, Computer Architecture: A
Quantitative Approach, 4th edition, Sept. 15, 2006
3X
??%/year
1000
52%/year
100
10
25%/year
 Sea change in chip
design: multiple “cores” or
processors per chip
1
1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
• VAX
: 25%/year 1978 to 1986
• RISC + x86: 52%/year 1986 to 2002
• RISC + x86: ??%/year 2002 to present
November 12th, 2009
Tessellation OS
Tessellation: 2
ManyCore Chips: The future is here
 Intel 80-core multicore chip
 80 simple cores
 Two floating point engines /core
 Mesh-like "network-on-a-chip“
 100 million transistors
 65nm feature size

“ManyCore” refers to many processors/chip
 64?

(Feb 2007)
128? Hard to say exact boundary
How to program these?
 Use
2 CPUs for video/audio
 Use 1 for word processor, 1 for browser
 76 for virus checking???

Something new is clearly needed here…
November 12th, 2009
Tessellation OS
Tessellation: 3
Parallel Processing for the Masses

Why is the presence of ManyCore a problem?
 Parallel
computing has been around for 40 years with
mixed results

Many researchers, several generations, widely varying approaches
 Parallel
computing has never become a generic software
solution (especially for client applications)
 Suddenly, parallel computing will appear at all levels of our
computation stack




Cellphones
Cars (yes, Bosch is thinking of replacing some of the 70 processors
in a high end car with ManyCore chips)
Laptops, Desktops, Servers…
Time for the computer industry to panic a bit???
 Perhaps
November 12th, 2009
Tessellation OS
Tessellation: 4
Why might we succeed this time?

No Killer Microprocessor to Save Programmers (No Choice)



New Metrics for Success (Different Criteria)





Whole industry committed, so more working on it
If future growth of IT depends on faster processing at same price (vs.
lowering costs like NetBook)
User-Interactive Applications Exhibit Parallelism (New Apps)


Perhaps linear speedup is not the primary goal
Real Time Latency/Responsiveness and/or MIPS/Joule
Just need some new killer parallel apps
vs. all legacy SW must achieve linear speedup
Necessity: All the Wood Behind One Arrow (More Manpower)


No one is building a faster serial microprocessor
For programs to go faster, SW must use parallel HW
Multimedia, Speech Recognition, situational awareness
Multicore Synergy with Cloud Computing (Different Focus)


Cloud Computing apps parallel even if client not parallel
Manycore is cost-reduction, not radical SW disruption
November 12th, 2009
Tessellation OS
Tessellation: 5
Outline


What is the problem (Did this already)
Berkeley Parlab




Space-Time Partitioning



RAPPidS goals
Partitions, QoS, and Two-Level Scheduling
The Cell Model



Structure
Applications
Software Engineering
Space-Time Resource Graph
User-Level Scheduling Support (Lithe)
Tessellation implementation



Hardware Support
Tessellation Software Stack
Status
November 12th, 2009
Tessellation OS
Tessellation: 6
ParLab: a Fresh Approach to Parallelism

What is the ParLAB?

A new Laboratory on Parallelism at Berkeley






Remodeled “open floorplan” space on 5th floor of Soda Hall
10+ faculty, some two-feet in, others collaborating
Funded by Intel, Microsoft, and other affilliate partners
Goal: Productive, Efficient, Correct, Portable SW for 100+ cores &
scale as core increase every 2 years (!)
Application Driven! (really!)
Some History

Berkeley researchers from many backgrounds started meeting in
Feb. 2005 to discuss parallelism



Circuit design, computer architecture, massively parallel computing,
computer-aided design, embedded hardware and software,
programming languages, compilers, scientific programming, and
numerical analysis
Considered successes in high-performance computing (LBNL) and
parallel embedded computing (BWRC)
Led to “Berkeley View” Tech. Report 12/2006 and new Parallel
Computing Laboratory (“Par Lab”)

Won invited competition form Intel/MS of top 25 CS Departments
November 12th, 2009
Tessellation OS
Tessellation: 7
Par Lab Research Overview
Personal Image Hearing,
Parallel
Speech
Health Retrieval Music
Browser
Design Patterns/Motifs
Composition & Coordination Language (C&CL)
C&CL Compiler/Interpreter
Parallel
Libraries
Efficiency
Languages
Parallel
Frameworks
Sketching
Static
Verification
Type
Systems
Directed
Testing
Autotuners
Dynamic
Legacy
Communication &
Schedulers
Checking
Code
Synch. Primitives
Efficiency Language Compilers
Debugging
OS Libraries & Services
with Replay
Legacy OS
Hypervisor
November 12th, 2009
Multicore/GPGPU
ParLab Manycore/RAMP
Tessellation OS
Correctness
Diagnosing Power/Performance
Easy to write correct programs that run efficiently on manycore
Tessellation: 8
Target Environment: Client Computing
Massive Cluster
Gigabit Ethernet

ManyCore + Mobile Devices + Internet

Lots of Computational Resources




Must enable massive parallelism (not get in the way)
Many (relatively) Limited Resources:

Power, I/O bandwidth, Memory Bandwidth, User patience…
Must use these as efficiently as possible
Services backed by vast Internet resources




Clusters
Information can be preserved elsewhere
Access to remote resources must be streamlined
Obvious use of ManyCore in Services – but this is not the real problem
Things we are willing to change:

Software Engineering, Libraries, APIs, Services, Hardware
November 12th, 2009
Tessellation OS
Tessellation: 9
Music and Hearing Application
(David Wessel)

Musicians have an insatiable appetite for
computation + real-time demands
More channels, instruments, more
processing,
more interaction!
 Latency must be low (5 ms)
 Must be reliable (No clicks!)

1.
Music Enhancer
Enhanced sound delivery systems for home
sound systems using large microphone and
speaker arrays
 Laptop/Handheld recreate 3D sound over ear
buds

2.
Hearing Augmenter

3.
Handheld as accelerator for hearing aid
Novel Instrument User Interface
Berkeley Center for New Music and Audio
Technology (CNMAT) created a compact
loudspeaker array: 10-inch-diameter
icosahedron incorporating 120 tweeters.
New composition and performance systems
beyond keyboards
 Input device for Laptop/Handheld

November 12th, 2009
Tessellation OS
10
Tessellation: 10
Health Application: Stroke Treatment
(Tony Keaveny)
 Stroke treatment time-critical, need
supercomputer performance in hospital
 Goal: First true 3D Fluid-Solid Interaction
analysis of Circle of Willis
 Based on existing codes for distributed
clusters
November 12th, 2009
Tessellation OS
Tessellation: 11
Content-Based Image Retrieval
(Kurt Keutzer)
Relevance
Feedback
Query by example
Similarity
Metric
Image
Database

1000’s of
images
Candidate
Results
Built around Key Characteristics of personal
databases
 Very large number of pictures (>5K)
 Non-labeled images
 Many pictures of few people
 Complex pictures including people, events,
and objects
November 12th, 2009
Final Result
Tessellation OS
places,
12
Tessellation: 12
Robust Speech Recognition
(Nelson Morgan)
 Meeting Diarist
 Laptops/
Handhelds at meeting
coordinate to create speaker
identified, partially transcribed
text diary of meeting
Use
cortically-inspired manystream spatio-temporal
features to tolerate noise
November 12th, 2009
Tessellation OS
13
Tessellation: 13
Parallel Browser
(Ras Bodik)

Goal: Desktop quality browsing on handhelds
 Enabled

by 4G networks, better output devices
Bottlenecks to parallelize
 Parsing,
Rendering, Scripting
Speedup
Slashdot (CSS Selectors)
50
45
40
35
30
25
20
15
10
5
0
2ms
84ms
1
2
3
4
5
6
7
8
Hardware Contexts
November 12th, 2009
Tessellation OS
14
Tessellation: 14
Parallel Software Engineering

How do we hope to tackle parallel programming?


Through Software Engineering and Control of Resources
Two type of programmers:

Productivity programmers (90% of programmers)


Efficiency programmers (10% of programmers)


Parallel programmers, extremely competent at handling parallel
programming issues
Target new ways to express software so that is can be
execute in parallel


Not parallel programmers, rather domain specific programmers
Parallel Patterns
System support to avoid “getting in the way” of the result


Parallel Libraries, Autotuning, On-the-fly compilation
Explicitly managed resource containers (Partitions)
November 12th, 2009
Tessellation OS
Tessellation: 15
Architecting Parallel Software with
Patterns (Kurt Keutzer/Tim Mattson)
Our initial survey of many applications brought out
common recurring patterns:
“Dwarfs” -> Motifs
 Computational patterns
 Structural patterns
Insight: Successful codes have a comprehensible
software architecture:
 Patterns give human language in which to
describe architecture
November 12th, 2009
Tessellation OS
Tessellation: 16
Motif (nee “Dwarf”) Popularity

(Red Hot / Blue Cool)
How do compelling apps relate to 12 motifs?
November 12th, 2009
Tessellation OS
17
Tessellation: 17
Architecting Parallel Software
Decompose Tasks/Data
Order tasks
Identify Data Sharing and Access
Identify the Key
Computations
Identify the Software
Structure
•Pipe-and-Filter
•Agent-and-Repository
•Event-based
•Bulk Synchronous
•MapReduce
•Layered Systems
•Arbitrary Task Graphs
• Graph Algorithms
• Dynamic programming
• Dense/Spare Linear Algebra
• (Un)Structured Grids
• Graphical Models
• Finite State Machines
• Backtrack Branch-and-Bound
• N-Body Methods
• Circuits
• Spectral Methods
November 12th, 2009
Tessellation OS
Tessellation: 18
Par Lab is Multi-Lingual

Applications require ability to compose parallel code
written in many languages and several different parallel
programming models




Correctness through all means possible





Let application writer choose language/model best suited to task
High-level productivity code and low-level efficiency code
Old legacy code plus shiny new code
Static verification, annotations, directed testing, dynamic checking
Framework-specific constraints on non-determinism
Programmer-specified semantic determinism
Require common spec between languages for static checker
Common linking format at low level (Lithe) not
intermediate compiler form

Support hand-tuned code and future languages & parallel models
November 12th, 2009
Tessellation OS
Tessellation: 19
Selective Embedded Just-In-Time
Specialization (SEJITS) for Productivity
(Armando Fox)



Modern scripting languages (e.g., Python and Ruby) have
powerful language features and are easy to use
Idea: Dynamically generate source code in C within the
context of a Python or Ruby interpreter, allowing app to
be written using Python or Ruby abstractions but
automatically generating, compiling C at runtime
Like a JIT but


Selective: Targets a particular method and a particular
language/platform (C+OpenMP on multicore or CUDA on GPU)
Embedded: Make specialization machinery productive by
implementing in Python or Ruby itself by exploiting key features:
introspection, runtime dynamic linking, and foreign function
interfaces with language-neutral data representation
November 12th, 2009
Tessellation OS
Tessellation: 20
Autotuning for Code Generation
(Demmel, Yelick)

Problem: generating optimal code
like searching for needle in haystack

Manycore  even more diverse

New approach: “Auto-tuners”
 1st generate program variations of
combinations of optimizations
(blocking, prefetching, …) and data
structures
 Then compile and run to
heuristically search for best code
for that computer
Examples: PHiPAC (BLAS), Atlas
(BLAS), Spiral (DSP), FFT-W (FFT)

November 12th, 2009
Tessellation OS
Search space for
block sizes
(dense matrix):
• Axes are block
dimensions
• Temperature is
speed
21
Tessellation: 21
Outline


What is the problem (Did this already)
Berkeley Parlab




Space-Time Partitioning



RAPPidS goals
Partitions, QoS, and Two-Level Scheduling
The Cell Model



Structure
Applications
Software Engineering
Space-Time Resource Graph
User-Level Scheduling Support (Lithe)
Tessellation implementation



Hardware Support
Tessellation Software Stack
Status
November 12th, 2009
Tessellation OS
Tessellation: 22
Services Support for Applications

What systems support do we need for new ManyCore
applications?


Should we just port parallel Linux or Windows 7 and be done with it?
Clearly, these new applications will contain:

Explicitly parallel components



Direct interaction with Internet and “Cloud” services



Potentially extensive use of remote services
Serious security/data vulnerability concerns
Real Time requirements



However, parallelism may be “hard won” (not embarrassingly parallel)
Must not interfere with this parallelism
Sophisticated multimedia interactions
Control of/interaction with health-related devices
Responsiveness Requirements

Provide a good interactive experience to users
November 12th, 2009
Tessellation OS
Tessellation: 23
PARLab OS Goals: RAPPidS





Responsiveness: Meets real-time guarantees



Good user experience with UI expected
Illusion of Rapid I/O while still providing guarantees
Real-Time applications (speech, music, video) will be assumed



Programs not completely assembled until runtime
User may request complex mix of services at moment’s notice
Resources change rapidly (bandwidth, power, etc)


Application-Specific parallel scheduling on Bare Metal partitions
Explicitly parallel, power-aware OS service architecture


Fully integrated with persistent storage infrastructures
Customizations not be lost on “reboot”



Untrusted and/or buggy components handled gracefully
Combination of verification and isolation at many levels
Privacy, Integrity, Authenticity of information asserted
Agility: Can deal with rapidly changing environment
Power-Efficiency: Efficient power-performance tradeoffs
Persistence: User experience persists across device failures
Security and Correctness: Must be hard to compromise
November 12th, 2009
Tessellation OS
Tessellation: 24
The Problem with Current OSs

What is wrong with current Operating Systems?

They do not allow expression of application requirements



They do not provide guarantees that applications can use




In a parallel programming environment, ideal scheduling can depend
crucially on the programming model
They do not provide sufficient Security or Correctness



They do not provide performance isolation
Resources can be removed or decreased without permission
Maximum response time to events cannot be characterized
They do not provide fully custom scheduling


Minimal Frame Rate, Minimal Memory Bandwidth, Minimal QoS from
system Services, Real Time Constraints, …
No clean interfaces for reflecting these requirements
Monolithic Kernels get compromised all the time
Applications cannot express domains of trust within themselves without
using a heavyweight process model
The advent of ManyCore both:


Exacerbates the above with a greater number of shared resources
Provides an opportunity to change the fundamental model
November 12th, 2009
Tessellation OS
Tessellation: 25
A First Step: Two Level Scheduling
Monolithic
CPU and Resource
Scheduling
Resource Allocation
And
Distribution
Two-Level Scheduling
Application Specific
Scheduling

Split monolithic scheduling into two pieces:

Course-Grained Resource Allocation and Distribution



Chunks of resources (CPUs, Memory Bandwidth, QoS to Services)
distributed to application (system) components
Option to simply turn off unused resources (Important for Power)
Fine-Grained Application-Specific Scheduling


Applications are allowed to utilize their resources in any way they see fit
Other components of the system cannot interfere with their use of
resources
November 12th, 2009
Tessellation OS
Tessellation: 26
Important Mechanism:
Spatial Partitioning


Spatial Partition: group of processors acting within hardware boundary


Boundaries are “hard”, communication between partitions controlled
Anything goes within partition


Some number of dedicated processors
Some set of dedicated resources (exclusive access)
Each Partition receives a vector of resources



Complete access to certain hardware devices
Dedicated raw storage partition
Some guaranteed fraction of other resources (QoS guarantee):


Memory bandwidth, Network bandwidth
fractional services from other partitions
November 12th, 2009
Tessellation OS
Tessellation: 27
Resource Composition
Secure
Channel
Device
Drivers

Secure
Channel
Individual
Partition
Component-based design at all levels:



Secure
Channel
Balanced
Gang
Applications consist of interacting components
Requires composable: Performance, Interfaces, Security
Spatial Partitioning Helps:

Protection of computing resources not required within partition




High walls between partitions  anything goes within partition
“Bare Metal” access to hardware resources
Shared Memory/Message Passing/whatever within partition
Partitions exist simultaneously  fast inter-domain communication


Applications split into mutually distrusting partitions w/ controlled
communication (echoes of Kernels)
Hardware acceleration/tagging for fast secure messaging
November 12th, 2009
Tessellation OS
Tessellation: 28
Space-Time Partitioning
Space

Spatial Partitioning Varies over Time



Partitioning adapts to needs of the system
Some partitions persist, others change with time
Further, Partititions can be Time Multiplexed



Time
Services (i.e. file system), device drivers, hard realtime partitions
Some user-level schedulers will time-multiplex threads within a partition
Global Partitioning Goals:



Power-performance tradeoffs
Setup to achieve QoS and/or Responsiveness guarantees
Isolation of real-time partitions for better guarantees
November 12th, 2009
Tessellation OS
Tessellation: 29
Another Look: Two-Level Scheduling

First Level: Gross partitioning of resources



Goals: Power Budget, Overall Responsiveness/QoS, Security
Partitioning of CPUs, Memory, Interrupts, Devices, other resources
Constant for sufficient period of time to:

Amortize cost of global decision making
Allow time for partition-level scheduling to be effective

Allows AutoTuning of code to work well in partition




Hard boundaries  interference-free use of resources for quanta
Second Level: Application-Specific Scheduling




Goals: Performance, Real-time Behavior, Responsiveness, Predictability
CPU scheduling tuned to specific applications
Resources distributed in application-specific fashion
External events (I/O, active messages, etc) deferrable as appropriate

Global/cross-app decisions made by 1st level
Justifications for two-level scheduling?


E.g. Save power by focusing I/O handling to smaller number of cores
App-scheduler (2nd level) better tuned to application


Lower overhead/better match to app than global scheduler
No global scheduler could handle all applications
November 12th, 2009
Tessellation OS
Tessellation: 30
It’s all about the communication

We are interested in communication for many reasons:




Communication represents a security vulnerability
Quality of Service (QoS) boils down message tracking
Communication efficiency impacts decomposability
Shared components complicate resource isolation:

Need distributed mechanism for tracking and accounting of resource
usage

E.g.: How do we guarantee that each partition gets a guaranteed fraction
of the service:
Application A
Shared File Service
Application B
November 12th, 2009
Tessellation OS
Tessellation: 31
Tessellation: The Exploded OS

Device drivers
(Security/Reliability)
 Network Services
(Performance)

Firewall
Virus
Large Compute-Bound Intrusion
Application
Monitor


And
Adapt
Real-Time
Application
Identity
Persistent
Storage &
File System
HCI/
Voice
Rec
Normal Components split
into pieces


Persistent Storage
(Performance, Security,
Reliability)
 Monitoring services

Video &
Window
Drivers


Device
Drivers

Biometric, GPS,
Possession Tracking
Applications Given
Larger Partitions

Tessellation OS
Performance counters
Introspection
Identity/Environment
services (Security)


November 12th, 2009
TCP/IP stack
Firewall
Virus Checking
Intrusion Detection
Freedom to use
resources arbitrarily
Tessellation: 32
Tessellation in Server Environment
Cloud
Storage
BW QoS
QoS
Guarantees
Network
Network
QoS
Large Compute-Bound
Network
QoS
Large
Compute-Bound
Application
Monitor
Network
QoS
Large
Compute-Bound
Application
Monitor
And
QoS
Large
Compute-Bound
Application
Monitor
And
Adapt
Application Adapt
Monitor
And
OtherAdapt
Large I/O-Bound
And
OtherAdapt
Large
I/O-BoundDevices
Application
Other
Large
I/O-BoundDevices
Application
Disk
Other
Large
I/O-Bound
Devices
Application
Persistent
Storage
&
Disk
Persistent
Storage & I/O Devices
Parallel
FileApplication
System
Disk
I/O
Drivers
Persistent
Storage
&
Parallel
File System
Disk
I/O
Drivers
Persistent
Storage &
Parallel
File System
I/O
Drivers
Parallel File System
Drivers
November 12th, 2009
Tessellation OS
Tessellation: 33
Outline


What is the problem (Did this already)
Berkeley Parlab




Space-Time Partitioning



RAPPidS goals
Partitions, QoS, and Two-Level Scheduling
The Cell Model



Structure
Applications
Software Engineering
Space-Time Resource Graph
User-Level Scheduling Support (Lithe)
Tessellation implementation



Hardware Support
Tessellation Software Stack
Status
November 12th, 2009
Tessellation OS
Tessellation: 34
Defining the Partitioned Environment

Cell: a bundle of code, with guaranteed resources, running
at user level




Has full control over resources it owns (“Bare Metal”)
Contains at least one address space (memory protection domain),
but could contain more than one
Contains a set of secured channel endpoints to other Cells
Interacts with trusted layers of Tessellation (e.g. the “NanoVisor”)
via a heavily Paravirtualized Interface



E.g. Can manipulate its address mappings but does not know what page
tables even look like
We think of these as components of an application or the OS
When mapped to the hardware, a cell gets:


Gang-schedule hardware thread resources (“Harts”)
Guaranteed fractions of other physical resources


Physical Pages (DRAM), Cache partitions, memory bandwidth, power
Guaranteed fractions of system services
November 12th, 2009
Tessellation OS
Tessellation: 35
Space-Time Resource Graph
Cell 1
Resources:
4 Proc, 50% time
1GB network BW
25% File Server

Cell 2
Cell 3
Lightweight
Protection Domains
Space-Time resource graph: the explicit instantiation of
resource assignments


Directed Arrows Express Parent/Child Spawning Relationship
All resources have a Space/Time component


Cell 3
E.g. X Processors/fraction of time, or Y Bytes/Sec
What does it mean to give resources to a Cell?



The Cell has a position in the Space-Time resource graph and
The resources are added to the cell’s resource label
Resources cannot be taken away except via explicit APIs
November 12th, 2009
Tessellation OS
Tessellation: 36
Implementing the Space-Time Graph

Partition Policy layer (allocation)
Allocates Resources to Cells based
on Global policies
 Produces only implementable spacetime resource graphs
 May deny resources to a cell that
requests them (admission control)


Partition Policy Layer
(Resource Allocator)
Reflects Global Goals
Mapping layer (distribution)
Makes no decisions
Space-Time Resource Graph
Time-Slices at a course granularity
(when time-slicing necessary)
 performs bin-packing like operation Mapping Layer (Resource Distributer)
to implement space-time graph
 In limit of many processors, no time
multiplexing processors, merely
distributing resources



Partition Mechanism Layer
Implements hardware partitions and
secure channels
 Device Dependent: Makes use of
more or less hardware support for
QoS and Partitions

November 12th, 2009
Tessellation OS
Partition Mechanism Layer
ParaVirtualized Hardware
To Support Partitions
Tessellation: 37
What happens in a Cell Stays in a Cell

Cells are performance and security isolated from all other cells

Processors and resources are gang-scheduled


Unpredictable resource virtualization does not occur



Message arrivals (along channels)
Page faults, timer interrupts (for user-level preemptive scheduling),
exceptions, etc
Cells start with single protection domain, but can request more as
desired



Example: no paging without linking a paging library
Cells can control delivery of all events


All fine-grained scheduling done by a user-level scheduler
Initial protection domain becomes primary
For now, protection domains are Address Spaces, but can be other things
as well
CellOS: A layer of code within a Cell that looks like a
traditional OS


Not required for all Cells!
On Demand Paging, Address Space management, Preemptive
scheduling of multiple address spaces (i.e. processes)
November 12th, 2009
Tessellation OS
Tessellation: 38
Scheduling inside a cell

Cell Scheduler can rely on:




Course-grained time quanta allowing efficient fine-grained use of
resources
Gang-Scheduling of processors within a cell
No unexpected removal of resources
Full Control over arrival of events


Application-specific scheduling for performance


Lithe Scheduler Framework (for constructing schedulers)
Systematic mechanism for building composable schedulers


Parallel libraries with completely different parallelism models can be
easily composed
Application-specific scheduling for Real-Time

Label Cell with Time-Based Labels. Examples:




Can disable events, poll for events, etc.
Run every 1s for 100ms synchronized to ± 5ms of a global time base
Pin a cell to 100% of some set of processors
Then, maintain own deadline scheduler
Pure environment of a Cell  Autotuning will return same
performance at runtime as during training phase
November 12th, 2009
Tessellation OS
Tessellation: 39
Example of Music Application
Music program
Audio-processing / Synthesis Engine
(Pinned/TT partition)
Input device
(Pinned/TT Partition)
Output device
(Pinned/TT Partition)
Preliminary
Time-sensitive
Network
Subsystem
GUI Subsystem
Network
Service
(Net Partition)
Graphical
Interface
(GUI Partition)
Communication with other
audio-processing nodes
Outline


What is the problem (Did this already)
Berkeley Parlab




Space-Time Partitioning



RAPPidS goals
Partitions, QoS, and Two-Level Scheduling
The Cell Model



Structure
Applications
Software Engineering
Space-Time Resource Graph
User-Level Scheduling Support (Lithe)
Tessellation implementation



Hardware Support
Tessellation Software Stack
Status
November 12th, 2009
Tessellation OS
Tessellation: 41
What would we like from the Hardware?

A good parallel computing platform (Obviously!)

Good synchronization, communication



Vector, GPU, SIMD



Measurement: performance counters
Caches: Give exclusive chunks of cache to partitions





Techniques such as page coloring are poor-man’s equivalent
Memory: Ability to restrict chunks of memory to a given partition

Partition-physical to physical mapping: 16MB page sizes?
High-performance barrier mechanisms partitioned properly
System Bandwidth
Power

Ability to put partitions to sleep, wake them up quicly
Fast messaging support




Can exploit data parallel modes of computation
Partitioning Support


On chip  Can do fast barrier synchronization with combinational logic
Shared memory relatively easy on chip
Used for inter-partition communication
DMA, user-level notification mechanisms
Secure Tagging?
QoS Enforcement Mechanisms



Ability to give restricted fractions of bandwidth
Message Interface: Tracking of message rates with source-suppression for QoS
Examples: Globally Synchronized Frames (ISCA 2008, Lee and Asanovic)
November 12th, 2009
Tessellation OS
Tessellation: 42
RAMP Gold:
FAST Emulation of new Hardware

RAMP emulation model for Parlab
manycore




Single-socket manycore target
Split functional/timing model, both
in hardware

Timing
State

Arch
State

Timing
Model
Pipeline
November 12th, 2009
Functional

Model
Pipeline
SPARC v8 ISA -> v9
Considering ARM model
Functional model: Executes ISA
Timing model: Capture pipeline
timing detail (can be cycle accurate)
Host multithreading of both
functional and timing models
Built for Virtex-5 systems (ML505
or BEE3)
Tessellation OS
Tessellation: 43
Tessellation Architecture
Library OS
Functionality
Application
Or
OS Service
Sched
Reqs.
Partition
Mechanism
Layer
(Trusted)
Partition
Scheduler
Configure Partition
Resources enforced by
HW at runtime
Partition
Resizing
Callback API
Res.
Reqs.
Partition
Allocator
Configure
HW-supported
Communication
Interconnect Message
Physical
Cache
Bandwidth
Passing
Memory
CPUs
Tessellation
Kernel
Partition
Management
Layer
Comm.
Reqs
Custom
Scheduler
Performance
Counters
Hardware Partitioning Mechanisms
November 12th, 2009
Tessellation OS
44
Tessellation: 44
Tessellation Implementation Status

First version of Tessellation


~7000 lines of code in NanoVisor layer
Supports basic partitioning



Network Driver and TCP/IP stack running in partition



Cores and caches (via page coloring)
Fast inter-partition channels (via ring buffers in shared memory, soon
cross-network channels)
Devices and Services available across network
Hard Thread interface to Lithe – a framework for constructing userlevel schedulers
Currently Two ports


4-core Nehalem system
64-core RAMP emulation of a manycore processor (SPARC)


Will allow experimentation with new hardware resources
Examples:
 QoS Controlled Memory/Network BW
 Cache Partitioning
 Fast Inter-Partition Channels with security tagging
November 12th, 2009
Tessellation OS
Tessellation: 45
Conclusion

Berkeley ParLAB




Application Driven: New exciting parallel applicatoins
Tackling the parallel programming problem via Software Engineering
Parallel Programming Motifs
Space-Time Partitioning: grouping processors & resources
behind hardware boundary
Focus on Quality of Service
 Two-level scheduling

1) Global Distribution of resources
2) Application-Specific scheduling of resources
Bare Metal Execution within partition
 Composable performance, security, QoS


Tessellation OS
Exploded OS: spatially partitioned, interacting services
 Components




NanoVisor: Partitioning Mechanisms
Policy Manager: Partitioning Policy, Security, Resource Management
OS services as independent servers
November 12th, 2009
Tessellation OS
Tessellation: 46